This invention relates to the interpolation of traffic information.
As compared with a traffic information system which collects traffic information from roadside sensors, a floating car system can collect traffic information over a broader area at a lower cost. However, the random routes and data collecting timings of floating cars lead to a spatial and temporal deficiency in the collected floating car data (hereafter referred to as FCD for brevity). Information display or route search in a car navigation system cannot be properly performed if there are such deficiencies in the collected traffic information. Therefore, it is necessary to interpolate the FCD if they are to be used for such applications.
A technique for imputing the traffic information collected by roadside sensors is disclosed in, for example, JP-A-7-129893. According to the artifice disclosed there, deficiency in traffic information on a certain road-link is interpolated with other traffic information obtained from other road-links located upstream or downstream of, or parallel to the certain road-link, that is, by using available geographical relationships among road-links. On the other hand, JP-A-2005-004668 discloses an interpolation method which uses only FCD and does not depend on such geographical relationships among road-links and which involves statistical processing of FCD. According to this disclosure, raw FCD are first statistically processed to serve as data corresponding to the road-links of interest, and the processed data are then temporarily stored. When real-time FCD can be collected, the real-time FCD are used. When real-time FCD cannot be collected, the previously stored, statistically processed FCD are used instead. Another simple interpolation technique is also known wherein until old FCD are replaced by new FCD, the old FCD continue to be supplied as interpolation information.
Further, JP-A-2005-004668 teaches an interpolation technique for the interpolation of FCD using spatial correlation on multiple road-links. According to this technique, principal component analysis is performed on the FCD collected in the past, and correlated FCD components on plural road-links are calculated to serve as the bases related to the traffic information for those plural road-links. And the road-links on which real-time FCD were not collected are interpolated by using the bases calculated from the road-links on which real-time FCD were collected, depending on the spatially correlated FCD components on multiple road-links.
However, these conventional Interpolation techniques have the following problems. The techniques disclosed in JP-A-7-129893 and JP-A-2005-004668 documents cannot perform interpolation depending on the spatial correlation on multiple road-links if the FCD missing rate for road-links is high. For example, even in the case where 100,000 floating cars are used all over Japan, the average refresh cycle of collecting FCD is nearly once an hour per road-link. When the thus collected data are used as traffic information distributed every 5 minutes, the spatial missing rate will reach a percentage not less than 90%. Accordingly, even if the interpolation of the road-links having missing traffic information by using the traffic information of neighboring links is attempted, such an attempt will fail because situations occur frequently where the traffic information of the neighboring links are all missing as well. If the interpolation of the road-links having missing traffic information is performed by using the traffic information on remote road-links, the precision in interpolation is very poor in an area where the connections among the road-links are complicated so that the traffic information obtained through interpolation becomes far different from the actual real-time traffic information. On the other hand, if the process of statistically treating the past FCD is used, the interpolation of FCD with a high rate of link data missing is indeed possible, but the statistically processed traffic information will not exactly reflect the real-time traffic information.
According to JP-A-2005-004668, the principal component analysis of the FCD collected in the past is performed without depending on the connections among road-links so that the correlated traffic data components on plural road-links are subjected to calculations to generate the bases which represent the traffic information on the plural road-links. Further, the weighting coefficients for the bases are calculated by projecting the vector representing the real-time FCD into the space subtended by the bases. Estimated traffic information on the plural road-links is calculated by the linear combination of these bases with the thus obtained weighting coefficients used as coefficients for the bases. The real-time traffic information of the road-links having missing FCD is interpolated with the estimated traffic information. However, if the spatial missing rate of road-link data is extremely high, the amount of the link data affecting the result of interpolation is insufficient and it may happen that the precision in the resulted interpolation is poor. Since traffic condition changes at any time for various causes, the link data on the neighboring links that affect the link data of the links subjected to interpolation also fluctuates with time. So, when the link data missing rate is extremely high, it is hardly possible that the link data on the neighboring links that affect the link data of the links subjected to interpolation were sufficiently collected. If the spatial interpolation is performed with very scarce spatial samples, using the technique disclosed in JP-A-2005-004668, the resulted precision becomes poor.
Further, for example, let it be assumed that ten bases selected arbitrarily from among the bases obtained by the principal component analysis of the past interpolated FCD are used for interpolation and that an area under investigation consists of one hundred road-links. If the link data missing rate is 95%, real-time FCD can be collected on only five road-links. Accordingly, the projection of the real-time FCD onto respective bases becomes impossible. In the case where the missing rate is 90%, the number of the road-links on which real-time FCD can be collected becomes the same as the number of the selected bases. Since, however, the road-links on which real-time FCD can be collected do not necessarily have strong correlation to the selected bases, the scarcity of samples may still lead to unstable outputs.
An example of traffic information system is disclosed in U.S. patent application publication No. 2006/0206256A1. An example of the interpolation method for traffic data is disclosed in “SPATIAL INTERPOLATION OF REAL-TIME FLOATING CAR DATA BASED ON MULTIPLE LINK CORRELATION IN FEATURE SPACE”, by Masatoshi Kumagai, et al., pp 1-6, ITS World Congress, 8-12 Oct. 2006″.